Re: [PATCH 5/6] clk: tegra: Add support for Tegra132 CAR clocks

[Peter De Schrijver <pdeschrijver@nvidia.com>]

On Wed, Jul 16, 2014 at 09:44:10AM +0200, Thierry Reding wrote:
> * PGP Signed by an unknown key
> 
> On Tue, Jul 15, 2014 at 06:24:35PM +0300, Peter De Schrijver wrote:
> > Tegra132 CAR supports almost the same clocks as Tegra124 CAR. This patch
> > deals with the small differences.
> > 
> > --
> > I'm not entirely sure why the soc_therm clock needs to be enabled on Tegra132,
> > but turning it off results in a system hang. I presume this might be because
> > of fastboot initializing soc_therm.
> > 
> > Signed-off-by: Peter De Schrijver <pdeschrijver@nvidia.com>
> > ---
> >  drivers/clk/tegra/clk-tegra124.c |   32 ++++++++++++++++++++++++++++++++
> >  1 files changed, 32 insertions(+), 0 deletions(-)
> > 
> > diff --git a/drivers/clk/tegra/clk-tegra124.c b/drivers/clk/tegra/clk-tegra124.c
> > index 80efe51..b857aab 100644
> > --- a/drivers/clk/tegra/clk-tegra124.c
> > +++ b/drivers/clk/tegra/clk-tegra124.c
> > @@ -1369,6 +1369,7 @@ static struct tegra_clk_init_table init_table[] __initdata = {
> >  	{TEGRA124_CLK_XUSB_HS_SRC, TEGRA124_CLK_PLL_U_60M, 60000000, 0},
> >  	{TEGRA124_CLK_XUSB_FALCON_SRC, TEGRA124_CLK_PLL_RE_OUT, 224000000, 0},
> >  	{TEGRA124_CLK_XUSB_HOST_SRC, TEGRA124_CLK_PLL_RE_OUT, 112000000, 0},
> > +	{TEGRA124_CLK_SOC_THERM, TEGRA124_CLK_PLL_P, 51000000, 0},
> >  	/* This MUST be the last entry. */
> >  	{TEGRA124_CLK_CLK_MAX, TEGRA124_CLK_CLK_MAX, 0, 0},
> >  };
> > @@ -1378,9 +1379,25 @@ static void __init tegra124_clock_apply_init_table(void)
> >  	tegra_init_from_table(init_table, clks, TEGRA124_CLK_CLK_MAX);
> >  }
> >  
> > +enum {
> > +	TEGRA124_CLK,
> > +	TEGRA132_CLK,
> > +};
> 
> I'd prefer this to be something like:
> 
> 	struct tegra_car_soc {
> 		bool has_ccplex_clk;
> 	};
> 
> 	static const struct tegra_car_soc tegra124_car_soc = {
> 		.has_ccplex_clk = false,
> 	};
> 
> 	static const struct tegra_car_soc tegra132_car_soc = {
> 		.has_ccplex_clk = true,
> 	};
> 
> > +static const struct of_device_id tegra_clock_of_match[] = {
> > +	{ .compatible = "nvidia,tegra124-car", .data = (void *)TEGRA124_CLK },
> 
> 	                                       .data = &tegra124_car_soc,
> 
> > +	{ .compatible = "nvidia,tegra132-car", .data = (void *)TEGRA132_CLK },
> 
> 	                                       .data = &tegra132_car_soc,
> 
> >  static void __init tegra124_clock_init(struct device_node *np)
> >  {
> >  	struct device_node *node;
> > +	const struct of_device_id *match;
> 
> 	const struct tegra_car_soc *soc;
> 
> > +	uintptr_t id;
> 
> > +	match = of_match_node(tegra_clock_of_match, np);
> > +	id = (uintptr_t)match->data;
> 
> 	soc = match->data;
> 
> >  
> >  	clk_base = of_iomap(np, 0);
> >  	if (!clk_base) {
> > @@ -1416,6 +1433,20 @@ static void __init tegra124_clock_init(struct device_node *np)
> >  	tegra_audio_clk_init(clk_base, pmc_base, tegra124_clks, &pll_a_params);
> >  	tegra_pmc_clk_init(pmc_base, tegra124_clks);
> >  
> > +	if (id == TEGRA132_CLK) {
> 
> 	if (soc->has_ccplex_clk) {
> 
> That's somewhat more explicit and avoids a lot of ugly casting.
> 

It also adds another struct + pointers to essentially store 1 bit. Which is why
I decided to go this route.

> > +		int i;
> > +
> > +		tegra124_clks[tegra_clk_cclk_g].present = false;
> > +		tegra124_clks[tegra_clk_cclk_lp].present = false;
> > +		tegra124_clks[tegra_clk_pll_x].present = false;
> > +		tegra124_clks[tegra_clk_pll_x_out0].present = false;
> > +
> > +		/* Tegra132 requires the soc_therm clock to be always on */
> > +		for (i = 0; i < ARRAY_SIZE(init_table); i++) {
> > +			if (init_table[i].clk_id == TEGRA124_CLK_SOC_THERM)
> > +				init_table[i].state = 1;
> 
> I wonder if we could do this someplace else. If we could, then we'd have
> the opportunity to make the init_table const.
> 

The easiest solution would be to turn on soc_therm for Tegra124 and Tegra132.
I don't think this would cause a measureable increase in power consumption.
If you're ok with this, this logic could just be removed. Another solution
would be to do an explicit clk_enable. PLL_P is already enabled so enabling
this clock, does not require PLL locking and could be done before udelay()
is available.

Cheers,

Peter.